{
  "id": 6377245,
  "title": "Muse, an AI personal agent launched by Meta",
  "url": "https://urgent.news/2026/09/09/muse-an-ai-personal-agent-launched-by-meta",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-09T03:20:29.000Z",
  "source": {
    "name": "The Economic Times",
    "slug": "the-economic-times",
    "url": "https://economictimes.indiatimes.com/news/new-updates/mark-zuckerbergs-meta-launched-muse-ai-agent-that-can-look-at-old-saved-instagram-reels-on-a-recipe-and-send-a-grocery-list-check-what-else-it-can-do/articleshow/133950653.cms"
  },
  "original_language": "en",
  "account": "Meta has launched an AI personal agent named Muse for the US market, beginning this week. The agent can remember details mentioned only once by a user and act on them without further reminders. Some of its capabilities include suggesting grocery lists from saved recipe reels, creating dinner party menus considering guests' dietary restrictions, handling day-to-day tasks like sending emails, booking travel, and completing forms, and executing larger goals by building plans, coordinating resources, and working behind the scenes until changes are needed or sensitive steps require approval.\n\nEach Muse agent operates on its own dedicated cloud computer, called Muse Secure VM, with Sentinel, another agent, monitoring and approving its actions before they reach the internet. Muse never sees users' passwords or payment details, as these are stored separately and used without visibility to the agent. For payments, Muse uses Link, a tool built by Stripe, providing purchase protections such as coverage for damaged or lost items, price drops, and fee-free returns. Each transaction utilizes a one-time virtual card to keep users' real card details private. Users decide which apps Muse can connect to and how much access each gets, with the ability to modify or revoke access anytime. Users can also instruct Muse to forget specific learned information and opt out of their interactions being used to train other Meta AI models. A more restricted version called Muse Confidential VM, which encrypts the entire virtual machine, including data and conversations, is scheduled for release later this year.\n\nThe launch of Muse has faced mixed results, with internal Meta staff flagging security concerns. Some internal testers reported impressive achievements, such as managing travel logistics for a honeymoon, while others raised serious issues, including exposure of a user's personal iCloud photo library during a toy identification task. Meta's chief technology officer, Andrew Bosworth, also experienced frequent logouts from the agent. Additionally, the launch comes at a time when similar patterns of unpredictable behavior have emerged in AI agents from other major tech companies like OpenAI and Anthropic. Within Meta, major technical and security incidents have increased by 40% compared to the previous year, with time spent by staff firefighting such incidents rising 70% over the same period. Muse is initially available through a dedicated app and WhatsApp, with support for Meta's AI glasses expected soon.",
  "summary": null,
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "TechCentral",
        "title": "Meta’s new AI agent can spend your money on your behalf",
        "url": "https://urgent.news/2026/09/09/metas-new-ai-agent-can-spend-your-money-on-your-behalf",
        "published": "2026-09-09T08:12:07.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}